Claude Code vs Codex: Which AI Coding Agent Should You Use?

Build with AI · Comparison

Both tools can read your project, write code, and run commands for you. The real question is which one fits the plan you already pay for and the way you like to work.

Claude Code and Codex are the two big AI coding agents right now. Claude Code comes from Anthropic. Codex comes from OpenAI. Both can look at your files, make changes, and run commands while you watch.

I build software with AI tools every week, and people ask me which one to pick. So here is a plain answer. I checked every price and feature below on the official pages on October 20, 2026.

What Claude Code and Codex are

Both tools are AI agents for code. You tell them what you want in plain words. They read your project, plan the change, edit files, and run commands like tests.

Anthropic describes Claude Code this way: "Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools." It runs in your terminal, your code editor, a desktop app, and the browser.

OpenAI describes the Codex command line tool like this: "Codex CLI is a coding agent from OpenAI that runs locally on your computer." Codex also works in your code editor, in the ChatGPT desktop app, and in the cloud, where tasks keep running after you close your laptop.

Price: what each one costs

Neither tool is sold on its own. You get each one through a plan you may already have. Here is what the official pages showed on October 20, 2026.

Claude Code pricing

• Free: $0, but Claude Code is not included. The plan table on claude.com shows Claude Code is not part of Free.
• Pro: $17 a month with the yearly discount ($200 billed up front), or $20 billed monthly. Claude Code is included.
• Max: from $100 a month, with 5x or 20x more usage than Pro.

Anthropic's pricing FAQ says it plainly: "Claude Code is included in all paid plans."

Codex pricing

• Free: $0. OpenAI's help center says "Codex is included across ChatGPT plans, including Free and Go." The ChatGPT pricing page calls it "Limited Codex access" on Free.
• Go: $8 a month for light coding tasks.
• Plus: $20 a month. This adds Codex on the web, in the command line, in your editor, and on iOS.
• Pro: from $100 a month, with plans at $100, $200, or $500.
• Business: $20 per user a month for 2 or more seats, billed yearly, on the openai.com Codex page.

One catch: Codex Cloud, where tasks run on OpenAI's computers, "is not included with Free or Go," per OpenAI's help center.

Usage limits: how much work you get

This is where most people get surprised. Both tools use limits that reset over time. Neither gives you a fixed number of messages.

Claude Code: Anthropic says "Every plan has usage limits that reset on a rolling five-hour session window, and paid plans add weekly limits on top." Your chats with Claude and your Claude Code work come out of the same pool. Pro gives "at least 5x more usage per 5-hour session than Free," and Max gives 5x or 20x the usage of Pro.

Codex: OpenAI's pricing docs show an estimated range of local messages per five hour period on Plus, and the range depends a lot on the model you pick. The same page says "Pro plans currently have no five-hour limit," and weekly limits may also apply. ChatGPT Work and Codex share usage.

When you hit a limit, both let you keep going for more money. Claude paid plans can turn on usage credits at API rates. ChatGPT Plus and Pro users can buy extra credits. Both tools also work with a pay as you go API key.

Setup: getting started with each

Setup is about the same for both. You install a small program, open your project folder, and sign in.

Claude Code setup

1. Open a terminal. On a Mac or Linux, Anthropic's docs give this command: curl -fsSL https://claude.ai/install.sh | bash
2. Go to your project folder and type claude.
3. Log in with your Claude account when it asks.

Windows has its own command in the docs. You need a Claude subscription (Pro, Max, Team, or Enterprise), a Claude Console account, or access through a supported cloud provider.

Codex setup

1. Open a terminal. OpenAI's GitHub page gives this Mac and Linux command: curl -fsSL https://chatgpt.com/codex/install.sh | sh
2. You can also install it with npm: npm install -g @openai/codex
3. Go to your project folder, type codex, and choose Sign in with ChatGPT.

Prefer not to use a terminal? Both have other doors. Claude Code has a desktop app and a web version. Codex works inside the ChatGPT desktop app and on the web. For a full walkthrough of the Claude side, see my guide on how to use Claude Code.

The real differences that matter

1. Project notes files

Both tools read a notes file in your project, so you do not repeat yourself. Claude Code uses CLAUDE.md files. Codex uses AGENTS.md. In Codex, the /init command makes one for you. Write your rules there once, like use plain CSS or run the tests before you finish.

2. Open source or not

The Codex CLI code is open source. Its GitHub page says "This repository is licensed under the Apache-2.0 License." Claude Code is not open source in the same way. Its license file says use is subject to Anthropic's Commercial Terms of Service. For most beginners this does not change much. For some companies it matters.

3. Cloud tasks

Codex Cloud runs tasks on OpenAI's computers, and OpenAI says cloud tasks "can continue while your computer is asleep." Claude Code also has a web version and a --cloud option in its docs, so you can start work locally and keep going on your phone. Both have moved fast here, so check the docs for your plan.

4. The models underneath

Claude Code runs on Anthropic's Claude models. Codex runs on OpenAI's GPT models. Model names change every few months, so I will not crown a winner on model quality. The test below will tell you more than any chart.

How to pick: the one afternoon test

Here is how I would decide if I were starting today. It costs little or nothing if you already have one plan.

1. Pick one small real task. For example: Add a contact form to my site that saves the name and email.
2. Make a copy of your project. Or use Git, so you can undo anything.
3. Give both tools the exact same prompt. Ask each one to make a plan first and wait for your OK.
4. Judge the result, not the hype. Did it work? Was the plan easy to follow? Did it ask good questions?
5. Check your usage page after. See how much of your limit the task used.

Keep the one that did better work for you. It is fine to use both. Some builders plan with one and review with the other. For a closer look at one of them, read my Claude Code review.

Who should pick which

My own summary, based on the official pricing pages checked October 20, 2026.

Comparing other tools too? See Cursor vs Claude Code and Claude Code pricing.